1: \begin{abstract}

2:     We provide a first  finite-particle convergence rate for Stein variational gradient descent (SVGD). 

3:     Specifically, 

4:     whenever the target distribution 

5:     is sub-Gaussian with a Lipschitz score, 

6:     %

7:     SVGD with $n$ particles and an appropriate step size sequence drives the kernel Stein discrepancy to zero at an order ${1/}{\sqrt{\log\log n}}$ rate.

8:     We suspect that the dependence on $n$ can be improved, and we hope that our explicit, non-asymptotic proof strategy will serve as a template for future refinements.

9: \end{abstract}

10: